Tutorials
Tutorial #1: Explainable Artificial Intelligence: A Genetic Programming Approach (Nov 17 · 1:30–3:30 PM · Room R215)
Presenter: Yi Mei, Professor, Victoria University of Wellington
Email:
Length: 110 mins
Abstract
Although machine learning has achieved great success in many real-world applications, it is criticised as usually behaving like a black box, and it is often difficult, if not impossible, to understand how the machine learning system makes the decision/prediction. This could lead to serious consequences, such as the accidents of the Tesla automatic driving cars, and biases of the automatic bank loan approval systems.
To address this issue, Explainable AI (XAI) is becoming a very hot research topic in the AI field due to its urgent needs in various domains such as finance, security, medical, gaming, legislation, etc. There have been an increasing number of studies on XAI in recent years, which improves the current machine learning systems from different aspects.
In evolutionary computation, Genetic Programming (GP) has been successfully used in various machine learning tasks including classification, symbolic regression, clustering, feature construction, and automatic heuristic design. As a symbolic-based evolutionary learning approach, GP has an inherent great potential to contribute to XAI, as a GP model tends to be interpretable. However, the interpretability in GP is not as straightforward as one expect it to be, and the models evolved by GP can still be huge and complex, thus less interpretable.
This tutorial will give a brief introduction on how one may achieve better model interpretability in XAI using GP. To this end, we will first briefly introduce XAI and GP. Then we will introduce the GP techniques/strategies that could lead to better model interpretability. We follow the common taxonomy of XAI and divide the techniques into intrinsic and post-hoc methods. In addition, we will also review some visualisation methods by/for GP to improve interpretability. The tutorial is concluded with some discussions on the current trend, challenges, and potential future research directions.
Tutorial #2: Agent-Based Modeling and Simulation: Design, Implementation, and Application in the AI Era (Nov 17 · 10:50 AM–12:20 PM · Room R215)
Presenter: Zhou He (贺舟), Professor, School of Economics and Management, University of Chinese Academy of Sciences
Email:
Length: 90 mins
Abstract
Situated at the intersection of complexity science and simulation, ABMS treats a system as a population of autonomous decision-making entities whose local rules generate outcomes that were never written into any central script. This tutorial equips participants to build such models deliberately rather than experimentally, and to hand them to others without loss. It opens with the conceptual vocabulary: what qualifies as an agent, how agent types relate to their instantiated individuals, why behavior splits into externally visible action and unobservable state, and how grid, continuous, geospatial, and network settings shape what agents can perceive and do. It then follows a model from its first sketch to its documented form: framing the question, sourcing and structuring data, choosing a platform, encoding rules, testing that the code behaves as intended and that the model behaves like the world, and recording every choice along the way, including a survey of the platforms available and the components they share. The centerpiece is VISA, a protocol that replaces free prose with eight interlocked tables covering agents, variables, sensing, internal functions, associated data, inputs and outputs, scheduling, and validation; nineteen executable rules make validity a checked property, while reusable AI skills carry a specification from authoring through checking to code generation across NetLogo, Python, and AnyLogic. Worked cases drawn from independently authored models show the workflow end to end, including where a proprietary library or missing data stops reproduction. The tutorial closes with the questions the field now has to answer: how LLM-driven agents are validated, how provenance is preserved, and where accountability stays with the modeler.
Tutorial #3: Evolutionary Feature Reduction for Machine Learning (Nov 17 · 8:30–10:30 AM · Room R215)
Presenters: Bach Hoai Nguyen, Senior Lecturer; Bing Xue, Professor, Victoria University of Wellington
Emails:
Length: 120 mins
Abstract
We are now in the era of big data, where vast amounts of high-dimensional data become ubiquitous in a variety of domains, such as social media, healthcare, and cybersecurity. When machine learning algorithms are applied to such high-dimensional data, they suffer from the curse of dimensionality, where the data becomes very sparse. Furthermore, the high-dimensional data might contain redundant and/or irrelevant features that blur useful information from relevant features. Feature reduction can address the above issues by building a smaller but more informative feature set.
Feature selection and feature construction are two main approaches of feature reduction. Feature selection aims to select a small subset of original (relevant) features. Feature construction aims to create a small set of new high-level (informative) features based on the original feature set. Although both approaches are essential data pre-processing steps, they are challenging due to their large search spaces and complex interactions between features. While exhaustive searches are impractical due to their intensive computation cost, traditional heuristic searches require less computational resources but can be trapped at local optima. Recently, evolutionary computation (EC) has been widely applied to achieve feature reduction because of its potential global search ability. Existing EC-based feature reduction approaches successfully reduce the data dimensionality while still improve the classification performance as well as the interpretability of the built models.
This tutorial firstly describes a general framework of feature reduction followed by the applications of feature reduction in real-world scenarios. Then, we will describe main EC techniques, including genetic algorithms, particle swarm optimisation, differential evolution, genetic programming, ant colony optimisation and evolutionary multi-objective optimisation (EMO) and show how they can be applied to address challenges in feature reduction. The effectiveness of EC-based feature reduction is illustrated through several applications including bioinformatics, image analysis and pattern classification, symbolic regression, and cybersecurity. The tutorial concludes with existing challenges for future research.
Tutorial #4: Advances in AI-Driven Data Generation: Methods and Applications (Nov 17 · 3:50–5:50 PM · Room R215)
Presenters: Chen Li, Associate Professor; Tadahiko Murata, Professor, The University of Osaka
Email:
Length: 110 mins
Abstract
Generative artificial intelligence has evolved from a technology for producing images and text into a powerful approach for creating structured, application-specific data. In scientific research, this shift is especially important because the space of possible solutions is often enormous, while collecting and evaluating real-world examples is costly. Molecular design offers a compelling illustration: conventional drug discovery relies heavily on searching existing compounds and iteratively optimizing candidates, whereas generative models can propose new molecular structures and guide exploration toward desired properties. Synthetic population generation and social simulation provide a complementary application area in which realistic artificial data can support analysis and simulation.
This tutorial introduces the foundations and recent advances of AI-driven data generation, with a particular emphasis on molecular generation for drug discovery. We first explain the motivations for generative approaches, the basic principles of variational autoencoders (VAEs), generative adversarial networks (GANs), diffusion models, and sequence-based generation, and the practical choices among molecular graph, SMILES, and SELFIES representations. We also discuss how to evaluate generated molecules in terms of validity, uniqueness, novelty, and relevant chemical properties.
The main part of the tutorial is organized around three progressively more constrained molecular design tasks. First, we examine de novo molecular generation and property optimization, using Transformer-based adversarial generation and reward-guided optimization to illustrate how models can move beyond reproducing a training distribution. Second, we introduce scaffold-based molecular generation, where a prescribed molecular core is retained while new decorations are generated and optimized. Third, we explore biologically informed molecular generation, in which gene expression profiles and complementary textual or biological information provide conditions for generating candidate molecules. Representative methods discussed in the slides include TenGAN, InstGAN, SpotGAN, GxVAEs, and ToDi, together with recent gene-expression-informed jointly controlled generation.
Throughout, the tutorial connects model architectures to their intended use cases, limitations, and evaluation considerations. We conclude by discussing the opportunities and open challenges of controllable, biologically meaningful, and experimentally useful AI-generated data.